Instrumental Variable Estimation in Theory and in Practice

Specialeforsvar: Amalie Doctor Jacobsen

Titel:  Instrumental Variable Estimation in Theory and in Practice:
Sensitivity of Estimators with Respect to Instrumental Variable Assumptions in Simulations, and Effect Estimation in a Clinical Trial with Compliance Issues

Abstract: This thesis work discusses how to handle estimation of causal effects in trials with treatment compliance issues, and in observational studies where an instrument may be identified. We consider both the binary treatment case and the bounded integervalued treatment case. Though, the binary treatment case is handled in most detail. In all cases a binary treatment assignment is used. The instrumental variable estimation framework is described and applied in some practical settings. In particular a number of simulations are carried out for examining what happens with the instrumental variable estimation when different instrumental variable assumptions are not satisfied, and for looking at how different missingness mechanisms influence the estimation. Finally the instrumental variable framework is used to analyse the clinical post-chemotherapy trial from Midtgaard et. al. [13]. A discussion is then made on how instrumental variable estimation may be used to supplement naive treatment effect estimates such as the intention to treat estimate, and on the appropriateness of instrumental variable estimation in practical settings.

Vejleder:  Anders Tolver
Censor:    Erik Thorlund Parner, Aarhus Universitet